No abstract

In this paper, we present an algorithm for multi person tracking in indoor surveillance systems based on tracking-by-detection approach. Convolutional Neural Networks (CNNs) for detection and tracking both are used. CNN Yolov3 has been utilized as detector. Person features extraction is performed based on modified CNN ResNet. Proposed architecture includes 29 convolutional and one fully connected layer. Hungarian algorithm is applied for objects association. After that object visibility in the frame is determined based on CNN and color features. For algorithm evaluation prepared videos that was labeled and tested using MOT evaluation metric. The proposed algorithm efficiency is illustrated and confirmed by our experimental results.

In this paper, a person tracking algorithm for indoor video surveillance is presented. The algorithm contains the following steps: person detection, person features formation, features similarity calculation for the detected objects, postprocessing, person indexing, and person visibility determination in the current frame. Convolutional Neural Network (CNN) YOLO v3 is used for person detection. Person features are formed based on H channel in HSV color space histograms and a modified CNN ResNet. The proposed architecture includes 29 convolutional and one fully connected layer. As the output, it forms a 128-feature vector for every input image. CNN model was trained to perform feature extraction. Experiments were conducted using MOT methodology on stable camera videos in indoor environment. Main characteristics of the presented algorithm are calculated and discussed, confirming its effectiveness in comparison with the current approaches for person tracking in an indoor environment. Our algorithm performs real time processing for object detection and tracking using CUDA technology and a graphics card NVIDIA GTX 1060.

No abstract

This paper discusses the algorithmic framework for image parking lot localization and classification for the video intelligent parking system. Perspective transformation, adaptive Otsu's binarization, mathematical morphology operations, representation of horizontal lines as vectors, creating and filtering vertical lines, and parking space coordinates determination are used for the localization of parking spaces in a~video frame. The algorithm for classification of parking spaces is based on the Histogram of Oriented Descriptors (HOG) and the Support Vector Machine (SVM) classifier. Parking lot descriptors are extracted based on HOG. The overall algorithmic framework consists of the following steps: vertical and horizontal gradient calculation for the image of the parking lot, gradient module vector and orientation calculation, power gradient accumulation in accordance with cell orientations, blocking of cells, second norm calculations, and normalization of cell orientation in blocks. The parameters of the descriptor have been optimized experimentally. The results demonstrate the improved classification accuracy over the class of similar algorithms and the proposed framework performs the best among the algorithms proposed earlier to solve the parking recognition problem.